cost_state_fn = BallSetFn(int(N/2), 0.25)
cost_fn = CostWrapper(cost_state_fn)
state_dim = 1
action_dim = 1
gen_model = GenerativeModel(trans_fn,
boundary,
cost_fn,
state_dim,
action_dim)
action_boundary = [(-1,1)]
discount = 0.95
problem = Problem(gen_model,
action_boundary,
discount)
return problem
if __name__ == '__main__':
parser = ArgumentParser(__file__,
'Generates a double integrator problem')
parser.add_argument('save_file',
metavar='FILE',
help='save file')
args = parser.parse_args()
problem = make_di_problem()
dump(problem,args.save_file)
mdp_obj = builder.build_mdp()
return (mdp_obj,discretizer)
if __name__ == '__main__':
parser = ArgumentParser(__file__,'Generates an MDP from continuous problem')
parser.add_argument('problem_in_file',
metavar='FILE',
help='problem file')
parser.add_argument('num_states',
metavar='N',
type=int,
help='number of states per dimension')
parser.add_argument('num_actions',
metavar='A',
type=int,
help='number of actions per dimension')
parser.add_argument('mdp_out_file',
metavar='FILE',
help='mdp save file')
parser.add_argument('disc_out_file',
metavar='FILE',
help='discretizer save file')
args = parser.parse_args()
problem=load(args.problem_in_file)
(mdp,disc) = make_uniform_mdp(problem,
args.num_states,
args.num_actions)
dump(mdp,args.mdp_out_file)
dump(disc,args.disc_out_file)
from solvers.value_iter import ValueIterator
from solvers import *
def solve_with_value_iter(mdp,thresh,max_iter):
iterator = ValueIterator(mdp)
solver = IterativeSolver(iterator)
term_conds = [ValueChangeTerminationCondition(thresh),
MaxIterTerminationCondition(max_iter)]
announce = [ValueChangeAnnounce()]
solver.termination_conditions.extend(term_conds)
solver.notifications.extend(announce)
solver.solve()
return iterator.get_value_vector()
if __name__ == '__main__':
parser = ArgumentParser(__file__,\
'Generates an MDP from continuous problem')
parser.add_argument('mdp_in_file',
metavar='FILE',
help='mdp file')
parser.add_argument('sol_out_file',
metavar='FILE',
help='solution out file')
args = parser.parse_args()
mdp = load(args.mdp_in_file)
sol = solve_with_kojima(mdp,1e-9,1e4)
dump(p, args.sol_out_file)
from mdp.state_functions import InterpolatedFunction
from mdp.solution_process import *
if __name__ == '__main__':
parser = ArgumentParser(__file__,\
'Form a policy from value function')
parser.add_argument('sol_in_file',
metavar='FILE',
help='solution in file')
parser.add_argument('disc_in_file',
metavar='FILE',
help='discretizer in file')
parser.add_argument('mdp_in_file',
metavar='FILE',
help='mdp in file')
parser.add_argument('v_fn_out_file',
metavar='FILE',
help='solution out file')
args = parser.parse_args()
p = load(args.sol_in_file)
disc = load(args.disc_in_file)
mdp = load(args.mdp_in_file)
# Build the q policy
(v,_) = split_solution(mdp,p)
v_fn = InterpolatedFunction(disc,v)
dump(v_fn,args.v_fn_out_file)